CN112949414A - Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image - Google Patents

Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image Download PDF

Info

Publication number
CN112949414A
CN112949414A CN202110153648.2A CN202110153648A CN112949414A CN 112949414 A CN112949414 A CN 112949414A CN 202110153648 A CN202110153648 A CN 202110153648A CN 112949414 A CN112949414 A CN 112949414A
Authority
CN
China
Prior art keywords
resolution
water body
image
wide
surface water
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202110153648.2A
Other languages
Chinese (zh)
Other versions
CN112949414B (en
Inventor
江威
庞治国
吕娟
杨昆
杨永民
付俊娥
路京选
李小涛
曲伟
李琳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
China Institute of Water Resources and Hydropower Research
Original Assignee
China Institute of Water Resources and Hydropower Research
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by China Institute of Water Resources and Hydropower Research filed Critical China Institute of Water Resources and Hydropower Research
Priority to CN202110153648.2A priority Critical patent/CN112949414B/en
Publication of CN112949414A publication Critical patent/CN112949414A/en
Application granted granted Critical
Publication of CN112949414B publication Critical patent/CN112949414B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/10—Terrestrial scenes
    • G06V20/13—Satellite images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/20—Image preprocessing
    • G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
    • G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
    • G—PHYSICS
    • G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
    • G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
    • G09B29/00—Maps; Plans; Charts; Diagrams, e.g. route diagram
    • G09B29/003—Maps
    • G09B29/004—Map manufacture or repair; Tear or ink or water resistant maps; Long-life maps
    • G—PHYSICS
    • G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
    • G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
    • G09B29/00—Maps; Plans; Charts; Diagrams, e.g. route diagram
    • G09B29/003—Maps
    • G09B29/005—Map projections or methods associated specifically therewith
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A90/00—Technologies having an indirect contribution to adaptation to climate change
    • Y02A90/30—Assessment of water resources

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Educational Administration (AREA)
  • Multimedia (AREA)
  • Educational Technology (AREA)
  • Business, Economics & Management (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Astronomy & Astrophysics (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Remote Sensing (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Image Processing (AREA)

Abstract

The invention discloses an intelligent earth surface water body drawing method for a wide-view high-resolution six-satellite image, which is characterized in that on the basis of selecting high-quality wide-view high-resolution six-remote-sensing image data, the geometric fine correction and image fusion method based on grid division is adopted to obtain the wide-view 2m resolution 8 wave band high-resolution six-fusion image data, the image multi-scale rapid segmentation is carried out based on principal component analysis and object facing combination, a characteristic classification vector is constructed on the scale of a segmented object, a water body/non-water body training sample with high space-time representativeness is selected based on a space grid, an object-facing deep neural network earth surface water body intelligent drawing model is constructed, and the automatic earth surface water body drawing of the wide-view high-resolution six-remote-sensing image is realized. The method mainly improves the efficiency and the precision of geometric fine correction, multi-scale segmentation and intelligent surface water extraction of the wide-view-field image, and has good application potential in flood monitoring, river and lake supervision and water ecological investigation.

Description

Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image
Technical Field
The invention belongs to the technical field of remote sensing image intelligent identification, and particularly relates to a design of an intelligent surface water body mapping method for a wide-vision high-resolution six-satellite image.
Background
The high-resolution surface water body mapping product is basic data of water resource investigation, water environment monitoring, water ecology evaluation and water disaster emergency. Since the implementation of a high-resolution earth observation system in China is a great special item, high-resolution space satellites from high-resolution one to high-resolution eight and the like are transmitted in sequence, and the requirement of high-resolution remote sensing data application in China is greatly met. Generally, the high-resolution image has a narrow imaging width, resulting in a long re-turn period, and for example, the high-resolution second satellite has a spatial resolution of 0.8m, a width of 45 km, and a theoretical re-turn time of 5 days. In order to improve the observation frequency of high-resolution satellites, a high-resolution six-satellite developed in China belongs to a wide-view low-orbit optical satellite and is provided with a 2-meter panchromatic/8-meter multispectral high-resolution camera and a 16-meter multispectral medium-resolution wide-width camera, wherein the width of the high-resolution camera is 90 kilometers, and the width of the wide-width camera is 800 kilometers, so that a rich data source is provided for developing large-scale, high-frequency and detailed surface water body mapping.
Aiming at optical remote sensing data, common surface water body mapping methods mainly comprise a single-waveband threshold value method, an inter-spectrum relation method, a water body index method, remote sensing supervision and classification and other methods. With the development of artificial intelligence and machine learning algorithms, the information of the surface water body can be effectively extracted by constructing a large number of sample training libraries and introducing artificial intelligence classification models. At present, a surface water body mapping method aiming at remote sensing data of sentinels No. 2, Landsat 8, Gao fen I and Gao fen II and the like exist, and a surface water body intelligent mapping method aiming at wide-view-field high-score six-grade satellite remote sensing images does not exist. Compared with other high-resolution series satellites, the high-resolution six-number wide-view image has the characteristics of large width, high spatial resolution, short recurrence period and the like, so that high-precision geometric correction and image fusion based on meshing are carried out aiming at the characteristics of the high-resolution six-number width, multi-scale rapid image segmentation based on principal component analysis and object-oriented combination is provided, characteristic waveband vectors are selected on the scale of a segmented object, representative training samples are selected based on a spatial grid, an object-oriented deep neural network earth surface water body intelligent drawing model is constructed, and automatic high-precision drawing of the earth surface water body of the wide-view six-number high-resolution remote sensing image is realized.
Disclosure of Invention
The invention aims to solve the problem that the prior art lacks an intelligent surface water body mapping method for wide-view high-resolution six-satellite remote sensing images, and provides an intelligent surface water body mapping method for wide-view high-resolution six-satellite remote sensing images, which is suitable for automatic high-precision surface water body mapping of the wide-view high-resolution six-satellite remote sensing images.
The technical scheme of the invention is as follows: an intelligent earth surface water body drawing method for wide-vision high-resolution six-satellite images comprises the following steps:
and S1, selecting wide-view high-resolution six-number remote sensing image data according to the time period to be monitored.
S2, preprocessing the wide-view-field high-resolution six-number remote sensing image data to obtain 2m resolution 8-waveband high-resolution six-number fusion image data.
And S3, performing multi-scale fast segmentation on the high-score six-number fusion image data by adopting a method of combining principal component analysis and object-oriented to obtain an image segmentation object.
S4, on the scale of the image segmentation object, constructing feature classification vectors and training samples, training and optimizing the deep neural network to obtain an object-oriented intelligent drawing model of the surface water body, and realizing automatic drawing of the wide-view high-resolution six-number remote sensing image surface water body.
Further, the wide-view-range high-resolution six-size remote sensing image data selected in step S1 includes 2m panchromatic/8 m multispectral high-resolution camera data and 16 m multispectral medium-resolution wide-width camera data.
Further, in step S1, the selection criteria of the wide-view-field high-resolution six-number remote sensing image data are:
(1) selecting a clear sky cloud-free image, wherein the surface water body area is free of cloud coverage;
(2) the image earth surface layer is clear, and no obvious aerosol covers;
(3) the image has no missing scanning lines, stripes, noise and abnormal pixels;
(4) the image is free of ice and snow.
Further, step S2 includes the following substeps:
s21, collecting 30m digital elevation model data and sentinel 2 data covering a wide-vision-field height-division six-size remote sensing image area, and taking the optical remote sensing data of the sentinel 2 with 10m spatial resolution as a reference image collection control point.
S22, dividing the wide-view-range high-resolution six-number remote sensing image into grids according to 3000 multiplied by 3000 pixels by adopting a grid division-based geometric fine correction method, extracting control point elevations from a digital elevation model by utilizing a frequency phase method and coordinates of reference image acquisition control points, manually removing image control points with errors larger than 2m, and reserving 20 high-precision control points in each grid.
And S23, optimizing rational function coefficients of the wide-view high-resolution six-number remote sensing image data based on the acquired high-precision control points, and realizing geometric fine correction on the high-resolution camera data and the wide-width camera data through a rational function model.
And S24, fusing the high-resolution camera data and the wide-width camera data after geometric fine correction by adopting a Panship image fusion method to obtain high-resolution six-number fusion image data with 2m resolution and 8 wave bands.
Further, step S3 includes the steps of:
s31, adopting a principal component analysis method to express the high-score six-number fusion image data as omega:
Ω＝[x1,x2,...,xn]
wherein xiThe two-dimensional pixel array representing the ith waveband of the high-resolution six-size fused image data indicates that the high-resolution six-size fused image data contains 8 pieces of image waveband information, wherein i is 1, 2.
S32, solving a covariance matrix K according to the high-score six-number fused image data omega:
Figure BDA0002933570390000031
and S33, performing eigenvalue decomposition on the high-resolution six-number fusion image data through the covariance matrix K, reducing the dimensionality of the high-resolution six-number fusion image data to 5 image wave bands, and taking the first 3 characteristic wave bands as images to be segmented.
S34, calculating the spatial heterogeneity phi of the high-resolution six-number fusion image data by adopting random seed points, and taking the spatial heterogeneity phi as a segmentation standard:
Φ＝φ1p+φ2q
wherein phi1And phi2All are weight values, p represents spectral heterogeneity, and q represents shape heterogeneity.
And S35, setting multi-scale dynamic segmentation parameters, dynamically adjusting a multi-scale segmentation threshold, and performing multi-scale dynamic segmentation on the segmented object according to the segmentation standard to obtain the image segmented object.
Further, in step S35, the multi-scale dynamic segmentation parameters are specifically set as: setting a segmentation scale of 150, a shape factor of 0.5 and a compactness factor of 0.3 for large-area and regular-shape surface water bodies in the segmentation object; aiming at small and medium areas and broken surface water bodies of the segmented objects, the segmentation scale is set to be 80, the shape factor is 0.3, and the compactness factor is 0.2.
Further, step S4 includes the following substeps:
and S41, respectively selecting water body samples and non-water body samples on the scale of the image segmentation object.
S42, constructing 11 characteristic wave bands based on the water body and non-water body object samples.
And S43, respectively extracting the average value of 11 characteristic wave bands as a training sample for each water body or non-water body object sample, training and optimizing the deep neural network to obtain an object-oriented intelligent surface water body drawing model, and realizing automatic drawing of the surface water body.
Further, the 11 characteristic bands constructed in step S42 include 8 image characteristic bands and 3 index characteristic bands, where the 8 image characteristic bands correspond to 8 image band information included in the top-ranked six fused image data, and the 3 index characteristic bands include a normalized vegetation index NDVI, a normalized water body index NDWI, and a SLOPE.
The calculation formula of the normalized vegetation index NDVI is as follows:
Figure BDA0002933570390000032
where ρ isnirInformation representing the characteristic band 4 of the image, predInformation indicating the image characteristic band 3.
The calculation formula of the normalized water body index NDWI is as follows:
Figure BDA0002933570390000033
where ρ isgreenInformation indicating the image characteristic band 2.
The SLOPE is calculated as:
Figure BDA0002933570390000041
wherein dz/dx represents the ratio of elevation z to the x-direction, and dz/dy represents the ratio of elevation z to the y-direction.
Further, the net function q (x) of the deep neural network in step S43 is:
Figure BDA0002933570390000042
where k is the total number of features, ωijIs a random initial weight value, bjIs a deviation, xiI is the ith characteristic band, i is the classification category.
The intelligent surface water drawing model judges whether each pixel category of the input image is a water body type or a non-water body type by constructing a classification discrimination function R (x):
Figure BDA0002933570390000043
where r (x) ═ 1 indicates that the class is a body of water, and r (x) ═ 0 indicates that the class is a body of water.
Further, in step S43, the SCRO function and the SGD function are used to optimize the weights of the neurons in the deep neural network, where the expression of the SCRO function Φ (α, β) is:
Figure BDA0002933570390000044
wherein x is a sample, n is a total amount of the sample, alpha is a real marked sample, the value is [0,1], beta is a probability value obtained by calculation through a flexible maximum activation function, and the calculation formula is as follows:
Figure BDA0002933570390000045
wherein e is an exponential function, xwaterVector matrix, x, representing water samplesnon-waterAnd representing a non-water body sample vector matrix.
The expression of the SGD function is:
Figure BDA0002933570390000046
wherein W' represents the weight after update, W represents the weight before update, B represents the bias, η represents the learning rate,
Figure BDA0002933570390000047
represents the gradient of the cost function J (·), X, Y are training sample pairs, and Z is the number of block samples.
The invention has the beneficial effects that:
(1) according to the invention, the wide-view-field high-resolution six-number remote sensing data in the research area needs to be selected according to the standard, the efficiency and the precision of geometric fine correction are improved by adopting a control point acquisition method based on grid division, and the provided method based on principal component analysis and object-oriented combination is beneficial to improving the efficiency of multi-scale segmentation and obtaining the image segmentation object.
(2) The object-oriented surface water body mapping model established based on the deep neural network can effectively solve the problem of incomplete water body boundary based on pixel classification, and realizes intelligent high-precision mapping of the surface water body.
(3) The method lays a foundation for carrying out large-scale high-frequency surface water body rapid mapping, and has good application potential in flood and drought emergency monitoring, river and lake dynamic supervision and water ecological remote sensing investigation.
Drawings
Fig. 1 is a flowchart of an intelligent mapping method for surface water bodies of wide-view high-resolution six-satellite images according to an embodiment of the present invention.
Fig. 2 is a flow chart of a method for intelligently mapping a surface water body of a wide-view high-resolution six-satellite image according to an embodiment of the present invention.
Fig. 3 is a drawing effect diagram of the earth surface water body of the wide-view high-resolution six-satellite remote sensing image according to the embodiment of the invention.
Detailed Description
Exemplary embodiments of the present invention will now be described in detail with reference to the accompanying drawings. It is to be understood that the embodiments shown and described in the drawings are merely exemplary and are intended to illustrate the principles and spirit of the invention, not to limit the scope of the invention.
The embodiment of the invention provides an intelligent mapping method for a wide-view high-resolution six-satellite image surface water body, which is shown in fig. 1-2 and comprises the following steps of S1-S4:
and S1, selecting wide-view high-resolution six-number remote sensing image data according to the time period to be monitored.
In the embodiment of the invention, suitable wide-view high-resolution six-number remote sensing image data are selected from a China resource satellite center website according to a time period to be monitored, the remote sensing image data comprise 2m panchromatic/8 m multispectral high-resolution camera data and 16 m multispectral medium-resolution wide camera data, and specific parameters are shown in table 1.
TABLE 1
Figure BDA0002933570390000051
Figure BDA0002933570390000061
In the embodiment of the invention, the selection standard of the wide-view-field high-resolution six-number remote sensing image data is as follows:
(1) selecting a clear sky cloud-free image, wherein the surface water body area is free of cloud coverage;
(2) the image earth surface layer is clear, and no obvious aerosol covers;
(3) the image has no missing scanning lines, stripes, noise and abnormal pixels;
(4) the image is free of ice and snow.
S2, preprocessing the wide-view-field high-resolution six-number remote sensing image data to obtain 2m resolution 8-waveband high-resolution six-number fusion image data.
The step S2 includes the following substeps S21-S24:
s21, collecting 30m Digital Elevation Model (DEM) data and sentinel 2 data covering a wide-view-field high-resolution six-number remote sensing image area, and taking the optical remote sensing data of the sentinel 2 with 10m spatial resolution as a reference image collection control point.
S22, considering that the imaging widths of the high resolution camera with the high resolution and the wide camera are large, a large number of redundant control points can be acquired through acquiring control points by the panoramic image. Therefore, in the embodiment of the invention, a grid division-based geometric precise correction method is adopted, the wide-view-range high-resolution six-number remote sensing image is divided into grids according to 3000 multiplied by 3000 pixels, the control point elevation is extracted from a digital elevation model by using a frequency phase method and coordinates of reference image acquisition control points, the image control points with the error larger than 2m are manually removed, and 20 high-precision control points are reserved in each grid.
S23, the rational function model is a geometric correction model commonly used for high-resolution images, rational function coefficients of wide-view high-resolution six-number remote sensing image data are optimized based on the acquired high-precision control points, and geometric fine correction is achieved on high-resolution camera data and wide-width camera data through the rational function model.
S24, the Panship image fusion method has a good effect of keeping image information, details and spectra, so that the Panship image fusion method is adopted to fuse high-resolution camera data and wide-width camera data after geometric fine correction in the embodiment of the invention to obtain 2 m-resolution high-resolution six-number fusion image data with 8 wave bands, space texture detail information is kept, and richer spectral information is also contained.
And S3, performing multi-scale fast segmentation on the high-score six-number fusion image data by adopting a method of combining principal component analysis and object-oriented to obtain an image segmentation object.
The step S3 includes the following steps S31 to S35:
s31, because the high-score six-number fusion image data contains 8 image wave band information, wherein the spectrum information of a plurality of image wave bands is redundant, and the calculation is time-consuming in multi-scale segmentation, the embodiment of the invention adopts a principal component analysis method to perform dimensionality reduction processing on the high-score six-number fusion image data so as to improve the efficiency of the multi-scale segmentation, and the high-score six-number fusion image data is expressed as omega:
Ω＝[x1,x2,...,xn]
wherein xiThe two-dimensional pixel array representing the ith waveband of the high-resolution six-size fused image data indicates that the high-resolution six-size fused image data contains 8 pieces of image waveband information, wherein i is 1, 2.
S32, solving a covariance matrix K according to the high-score six-number fused image data omega:
Figure BDA0002933570390000071
s33, decomposing the characteristic value of the high-resolution six-number fusion image data through the covariance matrix K, reducing the dimension of the high-resolution six-number fusion image data to 5 image wave bands, wherein the image pixel value is a value of projection conversion towards the direction of a principal component, and in order to improve the calculation efficiency, only the first 3 characteristic wave bands are taken as images to be segmented.
And S34, due to the fact that the difference between the shape and the type of the water body is large, the spectrum and the shape characteristics of the image can be effectively balanced through multi-scale segmentation, and accurate segmentation of the ground feature boundary is achieved. In the embodiment of the invention, the random seed points are adopted to calculate the spatial heterogeneity phi of the high-resolution six-number fusion image data, and the spatial heterogeneity phi is used as a segmentation standard, namely, images with the same spatial heterogeneity phi are segmented into one image block.
Φ＝φ1p+φ2q
Wherein phi1And phi2All are weight values, p represents spectral heterogeneity, and q represents shape heterogeneity. The spectral heterogeneity p is calculated according to the standard deviation of the image and the corresponding weight, and the shape heterogeneity q is divided according to the smoothness and compactness of the segmentation.
And S35, setting multi-scale dynamic segmentation parameters, dynamically adjusting a multi-scale segmentation threshold, and performing multi-scale dynamic segmentation on the segmented object according to the segmentation standard to obtain the image segmented object.
In the embodiment of the invention, the multi-scale dynamic segmentation parameters are specifically set as follows: setting a segmentation scale of 150, a shape factor of 0.5 and a compactness factor of 0.3 for large-area and regular-shape surface water bodies in the segmentation object; aiming at small and medium areas and broken surface water bodies of the segmented objects, the segmentation scale is set to be 80, the shape factor is 0.3, and the compactness factor is 0.2. By adjusting the multi-scale segmentation threshold value for multiple times, the boundary of the surface water element can be segmented into an image segmentation object more completely.
S4, on the scale of the image segmentation object, constructing feature classification vectors and training samples, training and optimizing the deep neural network to obtain an object-oriented intelligent drawing model of the surface water body, and realizing automatic drawing of the wide-view high-resolution six-number remote sensing image surface water body.
The step S4 includes the following substeps S41-S43:
and S41, respectively selecting water body samples and non-water body samples on the scale of the image segmentation object.
Because the wide-view high-resolution six-parameter remote sensing data has higher spatial resolution, the OpenCycleMap is used as the reference data in the embodiment of the invention, so that fine water body and non-water body samples can be selected, and the richness of the samples is ensured. The water body samples comprise types of lakes, rivers, pools, near-shore seawater, silt water bodies, aquaculture and the like, and the non-water body samples comprise types of ground objects such as farmlands, forests, cities, bare lands, bushes, mountain shadows, cloud shadows and the like.
S42, constructing 11 characteristic wave bands based on the water body and non-water body object samples.
In the embodiment of the invention, the 11 characteristic wave bands comprise 8 image characteristic wave bands and 3 index characteristic wave bands, the 8 image characteristic wave bands correspond to 8 image wave band information contained in the high-resolution six-number fusion image data, and the 3 index characteristic wave bands comprise a normalized vegetation index NDVI, a normalized water body index NDWI and a SLOPE SLOPE.
The calculation formula of the normalized vegetation index NDVI is as follows:
Figure BDA0002933570390000081
where ρ isnirInformation representing the characteristic band 4 of the image, predInformation indicating the image characteristic band 3.
The calculation formula of the normalized water body index NDWI is as follows:
Figure BDA0002933570390000082
where ρ isgreenInformation indicating the image characteristic band 2.
The SLOPE is calculated as:
Figure BDA0002933570390000083
wherein dz/dx represents the ratio of elevation z to the x-direction, and dz/dy represents the ratio of elevation z to the y-direction.
And S43, respectively extracting the average value of 11 characteristic wave bands as a training sample for each water body or non-water body object sample, training and optimizing the deep neural network to obtain an object-oriented intelligent surface water body drawing model, and realizing automatic drawing of the surface water body.
The deep neural network belongs to one of machine learning methods, and is used for deeply learning input features by constructing a plurality of hidden layer neuron network models, so that automatic extraction of a surface water body is realized. The embodiment of the invention constructs 11 characteristic wave bands, correspondingly has a random initial weight, and finally calculates a net function Q (x) through a plurality of hidden layer neural calculations, wherein the expression is as follows:
Figure BDA0002933570390000084
where k is the total number of features, ωijIs a random initial weight value, bjIs a deviation, xiI is the ith characteristic band, i is the classification category.
The intelligent surface water drawing model judges whether each pixel category of the input image is a water body type or a non-water body type by constructing a classification discrimination function R (x):
Figure BDA0002933570390000091
where r (x) ═ 1 indicates that the class is a body of water, and r (x) ═ 0 indicates that the class is a body of water.
In the embodiment of the present invention, an SCRO (software Cross-entry Objective) function and an sgd (storage Gradient decision) function are further required to optimize the weight of each neuron in the deep neural network, and finally, an optimal solution is obtained.
Wherein, the SCRO function continuously optimizes the weight of the deep neural network through reverse transfer by comparing the output with the real sample, and the expression is as follows:
Figure BDA0002933570390000092
wherein x is a sample, n is a total sample amount, alpha is a real mark sample, the value is [0,1], beta is a probability value obtained by calculation of a flexible maximum activation function, the function is that a water body and non-water body binary probability graph is calculated through each neuron, the output of each neuron is converted to 0-1, and the probability of the water body and the non-water body in the category is obtained, and the expression is as follows:
Figure BDA0002933570390000093
wherein e is an exponential function, xwaterVector matrix, x, representing water samplesnon-waterAnd representing a non-water body sample vector matrix. The SGD function updates the weights of the features in each iteration, each time calculated for 100 objects, with the expression:
Figure BDA0002933570390000094
wherein W' represents the weight after update, W represents the weight before update, B represents the bias, η represents the learning rate,
Figure BDA0002933570390000095
represents the gradient of the cost function J (·), X, Y are training sample pairs, and Z is the number of block samples.
In the embodiment of the invention, according to an artificial experiment, a deep neural network is provided with 4 hidden layers, 70% of samples are used for training a model, 30% of samples are used for optimizing the model, and the intelligent drawing result of the earth surface water body of the wide-view high-resolution six-size satellite image is shown in fig. 3, wherein a black area is the earth surface water body, so that the earth surface water body is accurately extracted, and the boundary profile of the extracted earth surface water body is complete.
And randomly generating 200 samples of the water body and the non-water body according to the surface water body extraction result, carrying out surface type artificial verification through a high-resolution satellite remote sensing image, constructing a confusion matrix, and further calculating the global precision and the Kappa coefficient according to confusion demonstration.
In the embodiment of the invention, aiming at the single-view wide-view high-resolution six-size satellite remote sensing image, the global precision is greater than or equal to 92% and the Kappa coefficient is greater than or equal to 0.8, so that the higher surface water body drawing precision can be realized.
It will be appreciated by those of ordinary skill in the art that the embodiments described herein are intended to assist the reader in understanding the principles of the invention and are to be construed as being without limitation to such specifically recited embodiments and examples. Those skilled in the art can make various other specific changes and combinations based on the teachings of the present invention without departing from the spirit of the invention, and these changes and combinations are within the scope of the invention.

Claims (10)

1. An intelligent earth surface water body drawing method for wide-vision high-resolution six-satellite images is characterized by comprising the following steps of:
s1, selecting wide-view high-resolution six-number remote sensing image data according to the time period to be monitored;
s2, preprocessing the wide-view-field high-resolution six-number remote sensing image data to obtain 2m resolution 8-waveband high-resolution six-number fusion image data;
s3, performing multi-scale fast segmentation on the high-score six-number fusion image data by adopting a method of principal component analysis and object-oriented combination to obtain an image segmentation object;
s4, on the scale of the image segmentation object, constructing feature classification vectors and training samples, training and optimizing the deep neural network to obtain an object-oriented intelligent drawing model of the surface water body, and realizing automatic drawing of the wide-view high-resolution six-number remote sensing image surface water body.
2. The method for intelligently mapping the earth surface water body of the wide-view high-resolution six-satellite image according to claim 1, wherein the wide-view high-resolution six-remote sensing image data selected in the step S1 includes 2m panchromatic/8 m multispectral high-resolution camera data and 16 m multispectral medium-resolution wide-width camera data.
3. The method for intelligently mapping the earth surface water body of the wide-view high-resolution six-satellite image according to claim 1, wherein the selection criteria of the wide-view high-resolution six-remote sensing image data in the step S1 are as follows:
(1) selecting a clear sky cloud-free image, wherein the surface water body area is free of cloud coverage;
(2) the image earth surface layer is clear, and no obvious aerosol covers;
(3) the image has no missing scanning lines, stripes, noise and abnormal pixels;
(4) the image is free of ice and snow.
4. The intelligent mapping method for the earth surface water body of the wide-view high-resolution six-satellite image according to claim 2, wherein the step S2 comprises the following substeps:
s21, collecting 30m digital elevation model data and sentinel 2 data covering a wide-vision-field height-division six-size remote sensing image area, and taking optical remote sensing data of sentinel 2 and 10m spatial resolution as a reference image collection control point;
s22, dividing the wide-view-range high-resolution six-number remote sensing image into grids according to 3000 multiplied by 3000 pixels by adopting a grid division-based geometric fine correction method, acquiring coordinates of control points by using a frequency phase method and a reference image, extracting control point elevations from a digital elevation model, manually removing image control points with errors larger than 2m, and reserving 20 high-precision control points in each grid;
s23, optimizing rational function coefficients of wide-view high-resolution six-number remote sensing image data based on the acquired high-precision control points, and realizing geometric fine correction on high-resolution camera data and wide-width camera data through a rational function model;
and S24, fusing the high-resolution camera data and the wide-width camera data after geometric fine correction by adopting a Panship image fusion method to obtain high-resolution six-number fusion image data with 2m resolution and 8 wave bands.
5. The intelligent mapping method for the earth surface water body of the wide-view high-resolution six-satellite image according to claim 1, wherein the step S3 comprises the following steps:
s31, adopting a principal component analysis method to express the high-score six-number fusion image data as omega:
Ω＝[x1,x2,...,xn]
wherein xiThe two-dimensional pixel array represents the ith wave band of the high-resolution six-number fused image data, wherein i is 1,2, the.
S32, solving a covariance matrix K according to the high-score six-number fused image data omega:
Figure FDA0002933570380000021
s33, decomposing the eigenvalue of the high-resolution six-number fusion image data through a covariance matrix K, reducing the dimensionality of the high-resolution six-number fusion image data to 5 image wave bands, and taking the first 3 eigen wave bands as images to be segmented;
s34, calculating the spatial heterogeneity phi of the high-resolution six-number fusion image data by adopting random seed points, and taking the spatial heterogeneity phi as a segmentation standard:
Φ＝φ1p+φ2q
wherein phi1And phi2All are weight values, p represents spectral heterogeneity, and q represents shape heterogeneity;
and S35, setting multi-scale dynamic segmentation parameters, dynamically adjusting a multi-scale segmentation threshold, and performing multi-scale dynamic segmentation on the segmented object according to the segmentation standard to obtain the image segmented object.
6. The method for intelligently mapping the earth surface water body of the wide-view high-resolution six-satellite image according to claim 5, wherein the multi-scale dynamic segmentation parameters in the step S35 are specifically set as follows: setting a segmentation scale of 150, a shape factor of 0.5 and a compactness factor of 0.3 for large-area and regular-shape surface water bodies in the segmentation object; aiming at small and medium areas and broken surface water bodies of the segmented objects, the segmentation scale is set to be 80, the shape factor is 0.3, and the compactness factor is 0.2.
7. The intelligent mapping method for the earth surface water body of the wide-view high-resolution six-satellite image according to claim 1, wherein the step S4 comprises the following substeps:
s41, respectively selecting water body samples and non-water body samples on the scale of the image segmentation object;
s42, constructing 11 characteristic wave bands based on the water body and non-water body object samples;
and S43, respectively extracting the average value of 11 characteristic wave bands as a training sample for each water body or non-water body object sample, training and optimizing the deep neural network to obtain an object-oriented intelligent surface water body drawing model, and realizing automatic drawing of the surface water body.
8. The method for intelligently mapping the earth surface water body of the wide-view high-resolution six-satellite image according to claim 7, wherein the 11 characteristic bands constructed in the step S42 include 8 image characteristic bands and 3 index characteristic bands, the 8 image characteristic bands correspond to 8 image band information included in the high-resolution six-fused image data, and the 3 index characteristic bands include a normalized vegetation index NDVI, a normalized water body index NDWI and a SLOPE;
the calculation formula of the normalized vegetation index NDVI is as follows:
Figure FDA0002933570380000031
where ρ isnirInformation representing the characteristic band 4 of the image, predInformation indicating a characteristic band 3 of the image;
the calculation formula of the normalized water body index NDWI is as follows:
Figure FDA0002933570380000032
where ρ isgreenInformation indicating the image characteristic band 2;
the calculation formula of the SLOPE is as follows:
Figure FDA0002933570380000033
wherein dz/dx represents the ratio of elevation z to the x-direction, and dz/dy represents the ratio of elevation z to the y-direction.
9. The method according to claim 7, wherein the net function Q (x) of the deep neural network in step S43 is:
Figure FDA0002933570380000034
where k is the total number of features, ωijIs a random initial weight value, bjIs a deviation, xiIs the ith characteristic wave band, i is the classification category;
the surface water intelligent mapping model judges whether each pixel category of the input image is a water body type or a non-water body type by constructing a classification discriminant function R (x):
Figure FDA0002933570380000035
where r (x) ═ 1 indicates that the class is a body of water, and r (x) ═ 0 indicates that the class is a body of water.
10. The method for intelligently mapping earth surface water body of wide-view high-resolution six-satellite image according to claim 7, wherein in step S43, the SCRO function and the SGD function are used to optimize the weight of each neuron in the deep neural network, and the expression of the SCRO function (α, β) is:
Figure FDA0002933570380000036
wherein x is a sample, n is a total amount of the sample, alpha is a real marked sample, the value is [0,1], beta is a probability value obtained by calculation through a flexible maximum activation function, and the calculation formula is as follows:
Figure FDA0002933570380000041
wherein e is an exponential function, xwaterVector matrix, x, representing water samplesnon-waterRepresenting a non-water body sample vector matrix;
the expression of the SGD function is as follows:
Figure FDA0002933570380000042
wherein W' represents the weight after update, W represents the weight before update, B represents the bias, η represents the learning rate,
Figure FDA0002933570380000043
represents the gradient of the cost function J (·), X, Y are training sample pairs, and Z is the number of block samples.
CN202110153648.2A 2021-02-04 2021-02-04 Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image Active CN112949414B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110153648.2A CN112949414B (en) 2021-02-04 2021-02-04 Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110153648.2A CN112949414B (en) 2021-02-04 2021-02-04 Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image

Publications (2)

Publication Number Publication Date
CN112949414A true CN112949414A (en) 2021-06-11
CN112949414B CN112949414B (en) 2022-06-03

Family

ID=76243731

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110153648.2A Active CN112949414B (en) 2021-02-04 2021-02-04 Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image

Country Status (1)

Country Link
CN (1) CN112949414B (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113408467A (en) * 2021-07-01 2021-09-17 中国科学院东北地理与农业生态研究所 Coastal culture pond intelligent extraction method based on Sentinel-2 satellite images and cloud platform
CN113837270A (en) * 2021-09-18 2021-12-24 广东人工智能与先进计算研究院 Target identification method, device, equipment and storage medium
CN116580320A (en) * 2023-05-25 2023-08-11 中国水利水电科学研究院 Large-scale intelligent remote sensing extraction method for artificial soil erosion disturbance range
CN116797855A (en) * 2023-08-22 2023-09-22 国网经济技术研究院有限公司 Method and device for detecting channel change of power transmission line based on satellite image data

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090226114A1 (en) * 2008-03-07 2009-09-10 Korea Aerospace Research Institute Satellite image fusion method and system
CN103383773A (en) * 2013-03-26 2013-11-06 中国科学院遥感与数字地球研究所 Automatic ortho-rectification frame and method for dynamically extracting remote sensing satellite image of image control points
CN112166693B (en) * 2012-06-29 2014-10-22 二十一世纪空间技术应用股份有限公司 Regional surface water resource remote sensing monitoring method based on small satellite
WO2018236032A1 (en) * 2017-06-23 2018-12-27 한국해양과학기술원 Device and method for delimiting/dividing seashore in order to cope with marine pollution accident
CN110427836A (en) * 2019-07-11 2019-11-08 重庆市地理信息和遥感应用中心(重庆市测绘产品质量检验测试中心) A kind of high-resolution remote sensing image Clean water withdraw method based on multi_dimension optimization
CN112052742A (en) * 2020-08-12 2020-12-08 河海大学 Semantic and pixel feature fused high-resolution binary remote sensing image water body extraction method

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20090226114A1 (en) * 2008-03-07 2009-09-10 Korea Aerospace Research Institute Satellite image fusion method and system
CN112166693B (en) * 2012-06-29 2014-10-22 二十一世纪空间技术应用股份有限公司 Regional surface water resource remote sensing monitoring method based on small satellite
CN103383773A (en) * 2013-03-26 2013-11-06 中国科学院遥感与数字地球研究所 Automatic ortho-rectification frame and method for dynamically extracting remote sensing satellite image of image control points
WO2018236032A1 (en) * 2017-06-23 2018-12-27 한국해양과학기술원 Device and method for delimiting/dividing seashore in order to cope with marine pollution accident
CN110427836A (en) * 2019-07-11 2019-11-08 重庆市地理信息和遥感应用中心(重庆市测绘产品质量检验测试中心) A kind of high-resolution remote sensing image Clean water withdraw method based on multi_dimension optimization
CN112052742A (en) * 2020-08-12 2020-12-08 河海大学 Semantic and pixel feature fused high-resolution binary remote sensing image water body extraction method

Non-Patent Citations (4)

* Cited by examiner, † Cited by third party
Title
TINA8101: "易康EPS2的使用（一）", 《CSDN BLOG.CSDN.NET/NECO8101/ARTICLE/DETAILS/103967350》 *
TINA8101: "易康EPS2的使用（一）", 《CSDN BLOG.CSDN.NET/NECO8101/ARTICLE/DETAILS/103967350》, 14 January 2020 (2020-01-14), pages 2 *
陈元鹏: "基于遥感数据的工矿复垦区分类与反演方法研究", 《中国优秀博硕士学位论文全文数据库(博士)基础科学辑》 *
陈元鹏: "基于遥感数据的工矿复垦区分类与反演方法研究", 《中国优秀博硕士学位论文全文数据库(博士)基础科学辑》, 15 September 2018 (2018-09-15), pages 2 - 3 *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113408467A (en) * 2021-07-01 2021-09-17 中国科学院东北地理与农业生态研究所 Coastal culture pond intelligent extraction method based on Sentinel-2 satellite images and cloud platform
CN113408467B (en) * 2021-07-01 2022-04-29 中国科学院东北地理与农业生态研究所 Coastal culture pond intelligent extraction method based on Sentinel-2 satellite images and cloud platform
CN113837270A (en) * 2021-09-18 2021-12-24 广东人工智能与先进计算研究院 Target identification method, device, equipment and storage medium
CN113837270B (en) * 2021-09-18 2022-08-30 广东人工智能与先进计算研究院 Target identification method, device, equipment and storage medium
CN116580320A (en) * 2023-05-25 2023-08-11 中国水利水电科学研究院 Large-scale intelligent remote sensing extraction method for artificial soil erosion disturbance range
CN116580320B (en) * 2023-05-25 2023-10-13 中国水利水电科学研究院 Large-scale intelligent remote sensing extraction method for artificial soil erosion disturbance range
CN116797855A (en) * 2023-08-22 2023-09-22 国网经济技术研究院有限公司 Method and device for detecting channel change of power transmission line based on satellite image data

Also Published As

Publication number Publication date
CN112949414B (en) 2022-06-03

Similar Documents

Publication Publication Date Title
CN112949414B (en) Intelligent surface water body drawing method for wide-vision-field high-resolution six-satellite image
CN110555446B (en) Remote sensing image scene classification method based on multi-scale depth feature fusion and migration learning
Chen et al. Spatially and temporally weighted regression: A novel method to produce continuous cloud-free Landsat imagery
CN112288647B (en) Remote sensing image cloud and shadow restoration method based on gating convolution
CN108428220B (en) Automatic geometric correction method for ocean island reef area of remote sensing image of geostationary orbit satellite sequence
CN113239830B (en) Remote sensing image cloud detection method based on full-scale feature fusion
Getirana et al. Improving hydrological information acquisition from DEM processing in floodplains
CN113705580B (en) Hyperspectral image classification method based on deep migration learning
CN111161199A (en) Spatial-spectral fusion hyperspectral image mixed pixel low-rank sparse decomposition method
CN113988147B (en) Multi-label classification method and device for remote sensing image scene based on graph network, and multi-label retrieval method and device
Ge et al. Improved semisupervised unet deep learning model for forest height mapping with satellite sar and optical data
CN111611960B (en) Large-area ground surface coverage classification method based on multilayer perceptive neural network
Cao et al. A full-level fused cross-task transfer learning method for building change detection using noise-robust pretrained networks on crowdsourced labels
CN114494821A (en) Remote sensing image cloud detection method based on feature multi-scale perception and self-adaptive aggregation
Yu et al. A deep transfer learning method for estimating fractional vegetation cover of sentinel-2 multispectral images
CN115908924A (en) Multi-classifier-based small sample hyperspectral image semantic segmentation method and system
Mao et al. Advances and prospects on estimating solar photovoltaic installation capacity and potential based on satellite and aerial images
Li et al. A new method for surface water extraction using multi-temporal Landsat 8 images based on maximum entropy model
CN113989612A (en) Remote sensing image target detection method based on attention and generation countermeasure network
CN116363526A (en) MROCNet model construction and multi-source remote sensing image change detection method and system
CN115797184A (en) Water super-resolution extraction model based on remote sensing image
CN115909077A (en) Hyperspectral image change detection method based on unsupervised spectrum unmixing neural network
Yamashkin et al. Classification of Metageosystems by Ensembles of Machine Learning Models
Li et al. Subpixel change detection based on improved abundance values for remote sensing images
CN111695530A (en) River water replenishing effect intelligent monitoring and evaluation method based on high-resolution remote sensing

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant